Robust modelling of the relationship between CD4 and viral load for complex AIDS data

Heavy Alcohol Use is Associated with Lower CD4 Counts among Russian Women Living with HIV: A Multilevel Analysis

Alcohol use remains prevalent among Russian women with HIV infection. Multilevel mixed effects models were used to estimate the association of heavy drinking and HIV outcomes among women (N = 250 at baseline; N = 207 at follow-up), aged 18-35, engaged in HIV care in Saint Petersburg. Alcohol use was assessed at baseline and 3 months by self-report and by the biomarker phosphatidylethanol (PEth). Overall, 35% of women were heavy drinkers, defined as women reporting ≥ 1 past-30-day heavy drinking episode (≥ 4 standard drinks on one occasion) or with PEth blood levels ≥ 80 ng/mL. Women who engaged in heavy drinking had an average 41 CD4 cells/mm³ (95% CI = - 81, - 2; z = - 2.04; P = 0.042) fewer than those who did not. Heavy drinking was associated with higher HIV symptom burden (IRR = 1.20; 95% CI = 1.05, 1.36; z = 2.73; P = 0.006) and suboptimal antiretroviral adherence (OR = 3.04; 95% CI = 1.27, 7.28; χ² = 2.50; P = 0.013), but not with viral load. Findings support the integration of alcohol treatment interventions as part of routine HIV care in Russia.

Double Poisson-Tweedie Regression Models

In this paper, we further extend the recently proposed Poisson-Tweedie regression models to include a linear predictor for the dispersion as well as for the expectation of the count response variable. The family of the considered models is specified using only second-moments assumptions, where the variance of the count response has the form μ+ϕμp $\mu + \phi \mu^p$, where µ is the expectation, ϕ and p are the dispersion and power parameters, respectively. Parameter estimations are carried out using an estimating function approach obtained by combining the quasi-score and Pearson estimating functions. The performance of the fitting algorithm is investigated through simulation studies. The results showed that our estimating function approach provides consistent estimators for both mean and dispersion parameters. The class of models is motivated by a data set concerning CD4 counting in HIV-positive pregnant women assisted in a public hospital in Curitiba, Paraná, Brazil. Specifically, we investigate the effects of a set of covariates in both expectation and dispersion structures. Our results showed that women living out of the capital Curitiba, with viral load equal or larger than 1000 copies and with previous diagnostic of HIV infection, present lower levels of CD4 cell count. Furthermore, we detected that the time to initiate the antiretroviral therapy decreases the data dispersion. The data set and R code are available as supplementary materials.